AI chips more suitable for the "Chinese physique": Both Xiaomi and Unitree are making efforts. NVIDIA Jetson Thor is now on sale, with a wholesale price of 20,000 yuan, but the delivery will take half a year.

Yesterday, Nvidia announced that its latest robot chip module, Jetson AGX Thor, is now officially on the market and will be sold as a developer kit for $3,499. It is reported that the performance of the Jetson Thor chip has been significantly improved compared to its previous - generation products. The company refers to this chip as the "robot brain," and customers can use it to develop robots.

The first batch of kits will be shipped next month and include the Jetson T5000 module, a reference carrier board with rich interfaces, an active radiator with a fan, and a power adapter. Nvidia will also simultaneously launch the Thor T5000 module for installation in robots ready for mass - production. If customers order 1,000 or more Thor chips at a time, the module will be sold at a wholesale price of $2,999 per piece (approximately 21,452 yuan). Currently, the delivery time for the Jetson AGX Thor development kit on the Arrow platform is 24 weeks.

Deepu Talla, vice president of Nvidia's Robotics and Edge AI division, said in a phone conference with reporters, "We don't produce robots, nor do we manufacture cars. However, we provide technical support for the entire industry through infrastructure computers and related software."

7.5 Times Faster Than the Previous Generation, and the Price Has Doubled

"There are millions of developers around the world working on robot systems that interact with and have an increasing impact on the physical world, and Jetson Thor is a product designed for them," said Jensen Huang, CEO of Nvidia.

It is reported that the Jetson Thor chip is built on the Blackwell graphics processor, which is the current - generation core technology of Nvidia. It is used in both its AI chips and game - specific chips. In recent years, Nvidia has continuously iterated and upgraded these "robot brains," and the performance of each generation of products has been improved compared to the previous one.

Nvidia claims that the AI computing power of the Jetson Thor chip is 7.5 times faster than that of the previous - generation product, NVIDIA Jetson Orin. Its energy efficiency has been increased by 3.5 times, CPU performance has been increased to 3.1 times, and memory capacity has been doubled, enabling real - time processing requirements at the edge. These performance improvements allow it to run generative AI models, including large - language models and visual models that can "interpret the surrounding environment" - which Nvidia says is crucial for humanoid robots.

However, the price of the NVIDIA Jetson AGX Orin 64G development kit is only $1,999.

In addition, the Jetson Thor chip is equipped with 128GB of memory and provides up to 2070 FP4 TFLOPS of AI computing power, which can easily run the latest AI models, all within a power range of 130W. "The key to developing powerful humanoid robots lies in our ability to run powerful AI models directly on the robot, enabling real - time learning and interaction," said Brett Adcock, founder and CEO of the humanoid robot development startup Figure AI Inc., previously.

Among the many demonstration cases shown by Nvidia for developers, a "hardware - in - the - loop" scenario is quite interesting. In this scenario, its Isaac GR00T N1 base model runs on the Jetson AGX Thor chip. This model is responsible for controlling a simulated robot, and this robot is virtually deployed through Isaac Sim (Nvidia's Isaac simulation platform) on the device. In the demonstration, this simulated robot is simulating a "nut pouring" task: it first picks up a beaker filled with nuts, pours one nut into a bowl, and then places the bowl on a scale to weigh it.

Wang Xingxing of Unitree Comes to "Endorse," Nvidia: More Suitable for Chinese Brands

According to Nvidia, currently, robot manufacturers such as Agility Robotics, Amazon, Meta, Boston Dynamics, United Imaging Healthcare, Wanjie Technology, Ubtech Robotics, Galaxy Universal, Unitree Technology, Zhongqing Robotics, and Zhiyuan Robotics are already using its Jetson Thor chip.

Among them, Agility Robotics has already used NVIDIA Jetson in its fifth - generation robot, Digit, and plans to use Jetson Thor as the computing core for the sixth - generation Digit. This change will further enhance Digit's real - time perception and decision - making capabilities, supporting increasingly complex AI capabilities and behavioral requirements. Digit has been commercialized and can perform logistics tasks such as stacking, loading, and palletizing in warehouse and manufacturing environments. Boston Dynamics, which has been deeply involved in the robot field for more than 30 years and has developed many industry - leading robots, is integrating Jetson Thor into its humanoid robot, Atlas. This will enable Atlas to carry computing power on the device that was previously only available on servers, accelerating edge - side AI workloads, high - bandwidth data processing, and supporting large - capacity memory.

Wang Xingxing, founder and CEO of Unitree Technology, said, "Jetson Thor brings a huge leap in computing power, empowering robots with stronger agility, faster decision - making, and a higher level of autonomy, which is crucial for robots to navigate and interact in the real world." Wang He, founder and CTO of Galaxy Universal, said, "After adopting Jetson Thor, our G1 Premium has achieved significant improvements in movement speed and action smoothness."

In addition to humanoid robots, Jetson Thor can also speed up various robot applications, including surgical assistance robots, intelligent tractors, delivery robots, industrial robotic arms, and visual AI agents, providing real - time inference capabilities for larger - scale and more complex AI models at the edge.

A research team from the Robotics Institute at Carnegie Mellon University uses NVIDIA Jetson to provide computing power support for autonomous robots. These robots can navigate in complex, unstructured environments and perform tasks such as medical triage and search and rescue. Sebastian Scherer, an associate research professor at the university and the head of the AirLab laboratory, predicts that after upgrading the team's existing NVIDIA Jetson AGX Orin system to the Jetson AGX Thor developer kit, it will not only improve the performance of AI models (including its award - winning robot edge perception model, MAC - VO) and enhance sensor fusion capabilities but also enable robot swarm experiments.

Nvidia claims that the Jetson Thor chip can also be applied in the field of autonomous vehicles, especially for Chinese automobile brands. The company named its automotive - specific chip Drive AGX. Although there are technical similarities between this type of chip and robot chips, the former runs an operating system called Drive OS - a system that has been specially optimized to better meet the needs of the automotive scenario. It is worth noting that the assisted - driving capabilities of Xiaomi's recently launched YU7 model are supported by the next - generation 4 - nanometer NVIDIA DRIVE AGX Thor™ in - vehicle computing platform, which can provide 700 TOPS of processing power, outperforming two combined Orin X chips.

Jensen Huang: Robots Are the Second - Largest Growth Opportunity

Some netizens commented on the release of Jetson AGX Thor, "Expanding from being limited to AI training to the fields of inference and robotics, Nvidia is continuously expanding its business scope. This is not just a product launch - it is a layout for the entire robot revolution."

Since 2014, Nvidia has successively launched several new robot chips. However, currently, the robotics business still accounts for a small proportion of its overall business, contributing only about 1% of the total revenue. The robotics business is still a small business for Nvidia.

However, this business is in a stage of rapid growth. Recently, Nvidia significantly adjusted its business disclosure method, grouping the automotive and robotics departments into the same business segment. The first - quarter earnings report for fiscal year 2026, disclosed in May this year, shows that the quarterly sales of this business unit were $567 million, accounting for about 1% of the total sales, but with a significant year - on - year increase of 72%.

Recently, Nvidia also invested in the robot startup Field AI. From the above layout, it is not difficult to see that Nvidia is currently heavily betting on the robotics field.

Previously, Jensen Huang said at the shareholders' meeting that the robotics field is the company's largest growth opportunity outside of artificial intelligence (AI) - it was the AI business that drove Nvidia's total sales to more than triple in the past two years. "We stopped seeing ourselves as just a chip company a long time ago. We are working towards a goal where, in the future, billions of robots, hundreds of millions of autonomous vehicles, and hundreds of thousands of robot factories can all be powered by Nvidia's technology," said Jensen Huang.

Reference Links:

https://blogs.nvidia.cn/blog/jetson-thor-physical-ai-edge/

https://blogs.nvidia.cn/blog/nvidia-blackwell-powered-jetson-thor-now-available-accelerating-the-age-of-general-robotics/

https://www.cnbc.com/2025/08/25/nvidias-thor-t5000-robot-brain-chip.html

This article is from the WeChat official account "AI Frontline." Compiled by Hua Wei. Republished by 36Kr with permission.